Trust Boundary

The conceptual line separating instructions and data an agent should trust (its own system prompt) from content it should treat skeptically (externally-sourced text), central to reasoning about prompt injection risk.